Whom to Help: Prosocial Behaviors and the Restoration of a Tarnished Identity

Moral Deviance and Restorative Events

Unethical behaviors, defined as “either illegal or morally unacceptable to the larger community” (Jones, 1991) can bring a host of negative consequences, upsetting the lives of wrongdoers in many ways and long after the misdeed occurred. Those who deviate from legal and moral norms often provoke condemnation, fear, and stereotypes about their dangerousness and general character (Conyers & Calhoun, 2015; Goode, 2015). Moral deviants face discrimination in the form of avoidance, lack of assistance, and support for punitive public policy that prevents access to resources (Cullen et al., 1985). Nonetheless, performing good deeds while abstaining from new moral transgressions can serve as an interactional resource to ameliorate the consequences of moral deviance (Foladare, 1969; Warren, 1980).

Psychological research on impression formation shows that good deeds can counteract the repercussions of wrongdoings. Levine and Schweitzer (2014) find that self-serving lies are more negatively perceived than “white lies.” Skowronski and Carlston (1992) and Riskey and Birnbaum (1974) find that pairing moral transgressions with good deeds improves impressions of the transgressor compared to when no pairing is done. Even though these findings suggest that prosociality results in more favorable perceptions of a wrongdoer, much less is known about the circumstances under which prosociality is more or less effective.

Affect control theory offers an impression-formation model that considers the effects of behaviors (e.g., moral transgressions and misdeeds) and many other features of the situation (e.g., emotions, settings, and the characteristics of the interactants). Previous studies of restorative actions using ACT have focused almost entirely on which emotional displays by victims and perpetrators affect impressions and sanctions (Robinson et al., 1994; Tsoudis, 2000; Tsoudis & Smith-Lovin, 1998, 2001; Zhao & Rogalin, 2017). More recently, Ramos et al. 2019 show that prosociality, or benefiting others at the expense of the actor (Simpson & Willer, 2008), can repair a moral transgressor’s image to a different extent depending on the perpetrator’s race and gender. Current research has not explored other situational elements that, according to ACT, should qualify the effect of prosociality on an actor’s impressions.

In this study, I broaden the existing literature by: (1) assessing how various degrees of prosociality can cast off the negative effects of unethical behaviors and (2) considering a previously ignored feature of restorative actions: who benefits from the good deed. In what follows, I describe ACT’s impression-formation model and derive hypotheses about the effect of prosociality and prosociality recipients on impressions of a wrongdoer.

Affect Control Theory

Among other aspects of social interaction, ACT models the affective impressions people form upon observing an event (see Heise, 2007; MacKinnon & Robinson, 2014; Robinson & Smith-Lovin, 2018 for comprehensive reviews of ACT). According to ACT, observers use cultural information about event actors, behaviors, and settings to form their impressions of the situation. Affect control theory (ACT) sees cultural knowledge about a concept as widely shared and measurable across three dimensions: evaluation, potency, and activity (Heise, 2007). Evaluation captures the goodness or badness of a concept, potency its powerfulness, and activity its liveliness (Osgood et al., 1957, 1975).

Affect control theory’s standard method to measure a concept’s meanings entails three semantic differential scales (one per dimension). Each scale ranges from −4 to +4 and its endpoints display the dimension’s negative and positive poles. For example, the evaluation scale displays “good” and “bad” at its positive and negative poles, respectively. ¹ The middle of the scale represents neutral (neither good nor bad) and is coded as 0. ² Individuals rate a concept by indicating where they think the concept lies on the scale. Fundamental sentiments represent the culturally shared affective meaning of a concept and are operationalized as the concept’s average evaluation, potency, and activity (EPA) ratings by individuals of the same culture.

Affect control theory compiles culture-specific fundamental sentiments in affective dictionaries. For example, a recent U.S. affective dictionary (Smith-Lovin et al., 2019) contains the fundamental sentiments of a mother (2.9, 2.43, and 1.35), indicating mothers are generally perceived as extremely good, quite powerful, and slightly active. The fundamental sentiments of a child in the same dictionary (2.21, −1.68, 2.39) indicate children are generally seen as quite good, quite powerless, and quite active. Affective dictionaries also include behaviors. For instance, the fundamental sentiment of “to kick someone” (−2.7, 0.91, 2.45) indicates an extremely bad, slightly powerful, and quite active behavior.

Affect control theory conceptualizes events as sequences of three elements: an actor, a behavior, and an object person to whom the behavior is directed (Heise, 1979, 2010). When an event occurs, the fundamental sentiments of the situation elements turn into transient impressions. For instance, observing a teacher kick a student turns the teacher’s fundamental positive evaluation into a negative transient impression. Transient impressions are also measured through semantic differential scales for each EPA dimension. I use this measurement model to capture impressions after unethical prosocial event sequences.

Affect control theory is based on impression-formation equations that predict the transient impressions a situation will evoke (e.g., Gollob, 1968; Heise & Smith-Lovin, 1981; Smith-Lovin, 1979; 1987a; 1987b). Impression-formation equations are the formal representation of impression-formation principles, which describe how fundamental sentiments turn into transient impressions after an event (Heise, 2007). INTERACT is computer software that uses ACT’s impression-formation equations to generate the expected transient EPA of each event element after simulated interactions (see (Schneider & Heise, 1995) for a complete description of INTERACT). ³ I use ACT impression-formation equations and INTERACT simulations to generate hypotheses about the effect of prosociality and the characteristics of prosociality beneficiaries.

I employ INTERACT simulations to predict transient actor evaluation after two scenarios that capture the theoretical features of interest. In a first scenario, a man steals from a man (a norm violating event). In a second scenario, a prosocial event follows the misdeed: a man steals from a man, and then a man hands cash to a firefighter, girl scout, mobster, or beggar.

INTERACT simulations take as input the fundamental sentiments that correspond to the actor(s), behavior(s), and object(s) of the event(s) and impression-formation equations. I employ gender-neutral fundamental sentiments from an ACT dictionary obtained using Amazon Mechanical Turk (MTurk) in 2015 (Smith-Lovin et al., 2019). Table 1 presents the fundamental sentiments of the actors, behaviors, and object persons involved in the simulations. As shown, a man, the actor in the events, is perceived as slightly good in the U.S. To steal something is perceived as extremely bad. The four object persons (firefighter, girl scout, mobster, and beggar) exhaust the four possible combinations of good, bad, powerful, and powerless identities. A firefighter and a girl scout are perceived as extremely and quite good, respectively. A beggar and a mobster are perceived as slightly and extremely bad, respectively. A firefighter and a mobster are perceived as extremely and quite powerful, respectively. A girl scout and a beggar both have negative scores in the potency dimension. The fundamental sentiments of “to hand cash to someone,” obtained from recently collected data by Lynn Smith-Lovin and collaborators indicate this behavior is considered quite good.

OPEN IN VIEWER

Table 1.

Transient Actor Evaluation by Simulation Scenario.

Simulation scenario	Object person	Object person fundamental sentiments			Transient actor evaluation males	Transient actor evaluation females
		E	P	A
“A man is”	—	—	—	—	.99	.99
“A man steals (−3.3, .53, .92) from a man”	man	.99	1.23	.69	−1.60	−1.39
“A man steals from a man. A man hands cash to (1.99, 1.62, .80) a [firefighter, girl scout, beggar, or mobster]”	firefighter	3.02	2.57	2.6	−.13	.22
	girl scout	2.32	−.32	1.3	−.13	.22
	beggar	−1.28	−2.64	−.40	−.41	−.10
	mobster	−2.74	2.27	1.84	−.66	−.38

An ideal study would include all affective dimensions and use identities that exhaust all combinations of good, bad, powerful, powerless, active, and inactive identities. However, some identity profiles are so scarce in the American culture that they are typified through inferred (e.g., has-been, bore) and even imaginary (e.g., zombie) entities. The literature recognizes the difficulty to identify terms with such diverse affective profiles (e.g., Robinson & Smith-Lovin, 1999). In this study, I focus on identities that most clearly demarcate evaluation and potency distinctions while acknowledging that the lack of all combinations of EPA profiles limits my ability to disentangle the effects of each dimension on transient actor evaluation.

I utilize the U.S.A 1978 impression-formation equations (available in INTERACT) to predict actor evaluation, the dependent variable of this study. The equations provide separate estimates for females and males. However, the estimates differ only in the intercept and not in the effects of interest for this study. Thus, I do not expect gender to qualify any of my hypotheses.

In this study, I focus on the following impression-formation principles: behavior effects, consistency effects, and congruency effects. Behavior effects capture the impact of prosociality after a misdeed. Behavior effects predict that bad behaviors reduce the actor’s evaluation while good behaviors increase the actor’s evaluation. Simulation results (in Table 1) show the transient man’s evaluation after he steals from a man is considerably lower than his fundamental evaluation. Across all object persons, prosociality leads to an increase in transient actor evaluation compared to when the man behaved unethically with no prosocial follow-up action. Therefore, I expect that:

H1

Actor prosociality will increase observers’ evaluation of the actor.

Consistency effects and congruency principles characterize the relationship between behavior evaluation and the affective profile of object persons. According to behavior-object evaluation consistency effects, being unkind toward a good object or kind toward a bad object reduces the actor’s evaluation. Congruency principles state how object potency qualifies behavior effects. The mercifulness effect holds that behaving nicely to weak objects increases the actor’s evaluation. Conversely, the sycophancy effect holds that behaving nicely to powerful objects reduces the actor’s evaluation.

INTERACT simulations illustrate how consistency effects and congruency principles influence actor evaluation. As shown in Table 1, transient actor evaluation varies according to the prosociality recipient. Consistent with ACT’s impression-formation principles, benefiting good object persons (girl scout and firefighter) leads to the greatest increase in actor evaluation, while benefiting a mobster (bad and powerful) leads to the smallest increase in actor evaluation, with beggar at the intermediate point, a pattern that reflects the contributions of both object evaluation and object potency on actor evaluation. Thus, I expect that:

H2

The positive effect of prosociality on observers’ evaluation of the actor will be greater when the prosocial behavior benefits a good object person compared to when it benefits a bad object person.

H3

The positive effect of prosociality on observers’ evaluation of the actor will be greater when the prosocial behavior benefits a powerless object person compared to when it benefits a powerful object person.

I test these predictions using two survey-vignette experiments, in which participants read a story describing events like those presented in the simulations.

Participants and Procedure

I recruited a convenience sample through MTurk. Studies indicate that MTurk samples are more diverse and of comparable or higher quality than undergraduate participant pools (Goodman et al., 2013; Paolacci & Chandler, 2014; Paolacci et al., 2010; Simons & Chabris, 2012; Weinberg et al., 2014). Since the study relies on shared U.S. cultural meanings, participants were screened to ensure they were born in the U.S., located in the U.S. at the time of the study, and were native English speakers. I also enabled tools from MTurk and Qualtrics that block individuals with IP addresses outside of the U.S. from entering the study.

The survey started with a tutorial for using semantic differential scales to provide EPA ratings. Slider scales ranged from −4 to +4 and contained nine equidistant markings, each labeled with an adverb: infinitely (at markings −4 and +4 of the scale), extremely (at −3 and +3), quite (at −2 and +2), slightly (at −1 and +1), and neutral (at 0). After completing the tutorial, participants were randomly assigned to read one of 12 vignettes.

In Study 1, I varied three vignette elements: level of prosociality, object evaluation, and object potency. The first experimental manipulation, prosociality, has three categories. In non-prosocial conditions, the man sees the object person but does not hand the person any cash. I included object persons in non-prosocial conditions to achieve parallelism. In partially prosocial conditions, the man hands half of the cash from the wallet. In completely prosocial conditions, the man gives away all the money from the wallet. The second and third experimental manipulations deal with characteristics of the prosociality recipient. I investigated two categories of object evaluation (good and bad) and two categories of object potency (powerful and powerless). The vignettes used the same identities from the simulations: firefighter, girl scout, mobster (described as a local mobster to make the character visually recognizable by participants), and beggar. The vignette used male pronouns when referring to the firefighter, mobster, and beggar and female pronouns when referring to the girl scout. The resulting experiment was a 3 × 2 × 2 factorial design with a total of 12 conditions. Table 2 lists the experimental manipulations and their operationalization in the vignette. Below is the vignette corresponding to the partially prosocial condition with the high-evaluation, low-potency object (girl scout):

“It’s a sunny day. A man is walking out of a grocery store when he comes across a wallet fat with cash. He stops, looks down and picks it up. While walking away from the store, the man takes the cash out of the wallet. After rounding a corner, the man notices a girl scout standing across the street. The man walks towards the girl scout and hands her half of the cash from the wallet. The man then walks away, disappearing down the street thinking about how he would spend the remaining cash from the wallet.”

OPEN IN VIEWER

Table 2.

Study 1-Experimental Manipulations.

Experimental manipulation	Categories	Vignette operationalization with girl scout as the object
Prosociality	Non-prosocial	After rounding a corner, the man notices a girl scout standing across the street. The man then walks away, disappearing down the street thinking about how he would spend the cash from the wallet.
	Partially prosocial	After rounding a corner, the man notices a girl scout standing across the street. The man walks towards the girl scout and hands her half of the cash from the wallet. The man then walks away, disappearing down the street thinking about how he would spend the remaining cash from the wallet.
	Completely prosocial	After rounding a corner, the man notices a girl scout standing across the street. The man walks towards the girl scout and hands her all the cash from the wallet. The man then walks away, disappearing down the street thinking about how the girl scout would spend the cash from the wallet.
Object evaluation	Bad	Local mobster, beggar
	Good	Firefighter, girl scout
Object potency	Powerless	Girl scout, beggar
	Powerful	Firefighter, local mobster

Note. Italicized words vary between non-prosocial and prosocial conditions. Underlined words vary between partially prosocial and completely prosocial conditions

The italicized elements indicate the manipulated vignetted elements; elements not italicized remained constant. The protagonist (a man) remained nameless to avoid biases arising from inferences about the protagonist’s race or socioeconomic status (Abascal, 2015). The location was also kept neutral to prevent setting influences on impressions (Woodward, 1997, pp. 21–22).

To be clear, the vignette design prioritized reproducing theoretical conditions of interest as opposed to achieving mundane realism. Some scenarios (e.g., handing money to a firefighter) might seem not to reflect real-life situations. However, this theory-driven study intends to empirically test theoretical principles and provide clues for understanding real-world situations. This goal is fundamentally different from that of many empirically-driven experiments seeking to document differences in a sample and generalize to the population (see Thye, 2014 and Zelditch, 1980 for further discussion of theory-driven vs. empirically-driven experiments). After reading the vignette, participants provided EPA ratings for “the man from the story.”

The survey included three active participation checks and two manipulation checks. Active participation checks asked respondents to select a particular option on survey items (e.g., “rate this as good”). Manipulation check items verified participants understood the story by asking: (1) whether the prosociality recipient saw the man pick up the wallet (correct answer: no); and (2) how the money was split between the man and the object person (the man kept all the money, the man kept half and the object person kept the other half, the man kept no money and the object person kept all of it, or none of the above; correct answer varied by condition).

A total of 592 individuals agreed to participate in the study. Out of these individuals, 204 failed active participation checks and were excluded immediately after failing the check (i.e., they did not complete the survey). A total of 388 participants completed the survey. I excluded data from 40 participants who failed manipulation checks: 26 participants failed the first manipulation check, 18 participants failed the second manipulation check, and four participants failed both. Thus, Study 1 included 348 participants. Appendix A includes models with control demographic variables to account for the possibility of non-random drops from failed manipulation checks. All findings hold when controlling for demographic variables.

Participants’ demographic characteristics resembled those of other MTurk studies. Participants’ age ranged from 20 to 74 years old, with a median of 33 years. 51.4% of the sample self-identified as female, 88.2% as white, 6.03% as black, and 7.18% as Hispanic. 50.3% of participants reported being liberal, 22.1% moderate, and 27.6% conservative.

Analytical Strategy

I perform OLS regressions to estimate the effects of interest on transient actor evaluation. The study’s dependent variable, transient actor evaluation, is observers’ rating of the story protagonist in the evaluation dimension, measured using a semantic differential scale (M = − .99, SD = 1.97, range = − 4.00 to 4.00). The experimental manipulations (prosociality, object evaluation, and object potency) are independent variables in the regression models. I treat prosociality as a categorical variable with three levels (non-prosocial, partially prosocial, and completely prosocial). Even though the object persons represent the four combinations of good-bad and powerful-powerless, their fundamental sentiments are asymmetrical; the fundamental evaluation of good object persons (E _firefighter = 3.02; E _{girl scout} = 2.32) does not exactly mirror the fundamental evaluation of bad object persons (E _beggar = −1.28; E _mobster = −2.74). Treating object evaluation and object potency as categorical variables considering these discrepancies would produce less accurate estimates. Therefore, I treat fundamental object evaluation and object potency as continuous variables in the regression models. I also include respondents’ gender as a covariate since ACT expects impressions to vary by observers’ gender. Gender was coded as female = 1, male = 0.

Note that the correlation coefficients for the fundamental sentiments of object persons are r = .17 for evaluation and potency, r = .96 for potency and activity, and r = .43 for evaluation and activity, suggesting confounding among affective dimensions. Therefore, care should be taken when interpreting results involving the potency dimension as these effects could be due to the activity profile of object persons.

Results

Table 3 presents the average transient actor evaluation obtained by experimental condition. Results show that prosociality tends to increase actor evaluation. When the man keeps all the money, his average transient evaluation ranges from −2.67 to −1.86. Keeping half of the money leads to more positive transient evaluation ratings (range = −1.91 to −.71). Transient actor evaluation becomes even more positive in completely prosocial conditions (range = −1.40 to 1.18). ACT simulations indicate transient actor evaluation is expected to be the most positive when a firefighter or a girl scout benefit from prosociality. However, descriptive results show helping the beggar results in much more positive evaluations than those obtained when helping other identities. This result is unexpected.

OPEN IN VIEWER

Table 3.

Average Actor Evaluation by Experimental Condition.

	Study 1
	Non-prosocial			Partially Prosocial			Completely Prosocial
	Avg	SD	N	Avg	SD	N	Avg	SD	N
Firefighter	−2.67	1.04	32	−.85	1.33	30	.39	2.10	37
Girl scout	−2.39	.74	23	−.83	2.14	29	.61	1.51	28
Mobster	−2.28	1.20	29	−1.91	1.03	25	−1.40	1.52	26
Beggar	−1.86	1.26	29	−.71	1.77	28	1.18	1.96	32

Study 2
	Partially prosocial			Completely prosocial
	Avg	SD	N	Avg	SD	N
Nurse	−1.11	1.57	23	.63	2.08	26
Janitor	−.80	1.88	23	.71	1.92	26
Drug dealer	−1.83	1.60	29	−1.79	1.59	26
Drug addict	−1.51	2.05	22	−.93	2.29	26
Fugitive	−1.88	1.45	25	−1.23	1.83	20

To formally assess results, Table 4 presents estimates from OLS regressions of transient actor evaluation on the experimental manipulations and their interactions. Models 1 to 3 present separate regression models for the effects of prosociality, prosociality by object evaluation, and prosociality by object potency. Model 4 presents a model that includes all terms.

OPEN IN VIEWER

Table 4.

OLS Regression for Transient Actor Evaluation.

	Transient actor evaluation
	Study 1 (N = 348)				Study 2 (N = 246)
	Model 1	Model 2	Model 3	Model 4	Model 1	Model 2	Model 3	Model 4
Partially prosocial (0 = non-prosocial)	1.27 (.21)***	1.20 (.21)***	1.29 (.22)***	1.24 (.21)***
Completely prosocial (0 = non-prosocial)	2.55 (.21)***	2.45 (.21)***	2.65 (.21)***	2.57 (.21)***
Completely prosocial (0 = partially prosocial)					.89 (.24)***	1.07 (.24)***	.90 (.24)***	1.05 (.24)***
Object evaluation	.11 (.04)**	−.03 (.06)	.12 (.04)**	−.06 (.06)	.31 (.05)***	.18 (.07)*	.30 (.05)***	.17 (.07)*
Partially prosocial (0 = non-prosocial)*Object evaluation		.19 (.09)*		.21 (.09)*
Completely prosocial (0 = non-prosocial)*Object evaluation		.25 (.09)**		.31 (.09)***
Completely prosocial (0 = partially prosocial)*Object evaluation						.25 (.10)**		.27 (.10)**
Object potency	−.22 (.04)***	−.22 (.04)***	−.14 (.07)*	−.10 (.07)	−.16 (.08)	−.17 (.08)*	−.17 (.12)	−.11 (.12)
Partially prosocial (0 = non-prosocial)*Object potency			−.03 (.10)	−.08 (.10)
Completely prosocial (0 = non-prosocial)*Object potency			−.19 (.09)*	−.27 (.10)**
Completely prosocial (0 = partially prosocial)*Object potency							.03 (.16)	−.12 (.17)
Female respondent	−.53 (.17)**	−.49 (.17)**	−.53 (.17)**	−.48 (.17)**	−.04 (.24)	−.05 (.24)	−.04 (.24)	−.05 (.24)
Constant	−1.97 (.17)***	−1.94 (.17)***	−2.01 (.17)***	−2.01 (.17)***	−1.22 (.20)***	−1.32 (.20)***	−1.23 (.20)***	−1.32 (.20)***
R-squared	.37	.38	.37	.40	.18	.20	.18	.20
Adjusted R-squared	.36	.37	.36	.38	.17	.19	.16	.18

Consistent with H1, results from Model 1 show a main effect of prosociality. Model 2 and Model 4 show significant interaction effects between prosociality and object evaluation. Figure 1 (Panel A) plots the predicted transient actor evaluation (using Model 4) at different values of object evaluation for each prosociality condition while holding object potency at its mean and respondent gender at male. The plot shows the effect of prosociality is more pronounced as object evaluation increases. Overall, these results support H2.OPEN IN VIEWER

Model 3 and Model 4 show a statistically significant interaction between prosociality and object potency. Figure 1 (Panel B) plots the predicted transient actor evaluation (using Model 4) at different values of object potency for each prosociality condition while holding object evaluation at its mean and respondent gender at male. The figure shows the effect of prosociality becomes more pronounced as object potency decreases. Finally, all models show female respondents provided lower transient actor evaluation ratings.

Study 1 Summary and Discussion

Results suggest prosociality after an unethical behavior can help mitigate the devaluation of its actor (H1). The effect of prosociality is qualified by the affective profile of beneficiaries in a manner consistent with H2 and H3. Descriptive results also indicate unexpectedly positive actor ratings when helping a beggar. The unexpected results could be due to the denotative (rather than affective) meanings of the identity beggar. Giving money to beggars might be perceived as not only acceptable, but also expected and customary, positively affecting perceptions of those performing this action. This type of norm or expectation is not currently captured by ACT’s impression-formation model, which only focuses on affective meanings.

The high transient evaluation the man obtained when helping a beggar could influence Study 1 estimates. Thus, Study 2 further tests H2 and H3 with a new set of identities that includes two (instead of one) identities perceived as bad and powerless. Testing H1, H2, and H3 on additional and somewhat different sets of EPA profiles also allows exploring the generalizability and robustness of Study 1’s findings.

Participants and Procedure

Recruitment, screening, and participation procedures were the same as for Study 1. A total of 422 individuals agreed to participate in the study. Out of these individuals, 145 failed active participation checks and, thus, were immediately excluded from the study. 277 participants completed the study. I excluded data from 31 participants who failed one or both manipulation checks: 16 participants failed the first manipulation check, 17 participants the second, and two participants both. Study 2 included 246 responses. The final sample was 46% female, 86% white, 6% black, and 5% Hispanic. Fifty-five percent of participants reported being liberal, 25% moderate, and 20% conservative. The median age was 33 years old.

Study 2 utilized the same core vignette and procedure as Study 1. Study 2 employed a 2 × 2 × 2 factorial design and differed from Study 1 in two ways. First, Study 2 only includes partially prosocial and completely prosocial conditions. Second, I manipulated the perceived object evaluation and object potency with a new set of identities with the following EPA profiles: nurse (3.00, 1.53, 1.82), janitor (1.52, −1.06, .69), drug dealer (−2.87, 1.23, 1.22), drug addict (−2.55, −2.42, .11), and fugitive (−2.54, −.84, 1.01). That is, Study 2 includes two identities perceived as bad and powerless. The vignette used male pronouns when referring to the janitor, drug dealer, drug addict, and fugitive and female pronouns when referring to the nurse. I follow the same analytical strategy utilized in Study 1 and include models with control demographic variables to account for the possibility of non-random drops from failed manipulation checks (see Appendix A). All findings hold when controlling for demographic variables.

Results

Table 3 presents the mean actor evaluation obtained in each experimental condition. Consistent with H1, actor evaluation became more positive the more prosocial the man’s behavior across all object persons. Consistent with expectations, the man obtained the most positive ratings when he handed all the money to the high-evaluation, low-potency object person (janitor). The man obtained the most negative ratings when benefiting a fugitive (in partially prosocial conditions) and a drug dealer (in completely prosocial conditions). The man received intermediate ratings when helping a nurse or a drug addict. Taken together, descriptive results indicate that both object evaluation and object potency qualified the effect of prosociality on actor evaluation in a manner consistent with H2 and H3.

Table 4 presents estimates of OLS regressions for transient actor evaluation on the experimental manipulations and their interactions. Consistent with H1, results from Model 1 show a main effect of prosociality. Results also show a significant interaction between prosociality and object evaluation. Figure 1 (Panel C) plots this interaction (by predicting transient actor evaluation using Model 4). The figure shows that behaving completely rather than partially prosocially had a more pronounced positive effect on actor evaluation as values of object evaluation increase. This evidence supports H2.

Models 3 and 4 for Study 2 show no significant interaction between prosociality and object potency. Figure 1 (Panel D) plots this interaction (by predicting transient actor evaluation using Model 4). The figure shows that behaving completely rather than partially prosocially had a more pronounced positive effect on actor evaluation as values of object potency decrease. Nonetheless, the effect did not attain statistical significance. Thus, results from Study 2 do not support H3.

Study 2 Summary and Discussion

Through Study 2, I conduct an additional test of the effects of the perceived goodness-badness and powerfulness-powerlessness of object persons in the context of unethical prosocial events. Study 2 aims to confirm whether Study 1’s findings hold when other identities receive the money. Results from Study 2 provide additional support for H1, which holds that observers will evaluate an actor more positively the more prosocial the behavior. Study 2 results provide evidence for H2, but do not align with H3.